An object-based scene-graph description of a scene facilitates data transmission because each object is represented by a cartoon-type model with multi-resolution capabilities. For instance, despite data transmission under strong limitations due to limited emitter and receiver performance, a model can be retrieved at any stage of the transmission.

Scene Segmentation

Scene segmentation aims to decompose a scene into 3D objects organized in an object-based scene-graph where objects are associated with their 2D texture image(s). A scene segmentation process breaks physical links between vertices of the initial mesh representation of the scene. By principle, segmentation decisions have to be made from all the spectral channels inherent to multi-modal data. For instance, geometry information is one of most important factors in the decision. Objects with different signatures in the modality space can also significantly improve the efficiency of the segmentation process.

Multi-Resolution Data Analysis

Multi-resolution analysis allows multi-resolution information extraction from geometrical information along with the associated textures. This process analyzes the data in a multi-resolution fashion to be used in other processes, such as data segmentation and data reduction. Edges and other geometrical information can be detected at different scales. The multi-resolution process, based on the wavelet transform for example, does not only analyze 2D data, such as range ../images or texture ../images, but also 3D meshes.

Data Reduction

Data reduction refers to the process of reducing the amount of information in a data set by preserving the details necessary to recognize a scene at any factor of reduction. By definition, the recognition phase is subjective because it is often user-dependent and sometimes task-dependent. However, geometrical errors between the initial and reduced model can be minimized. User needs can be taken into account in the reduction process through a multi-resolution analysis of the data. After extracting multi-resolution information, the user can select the most relevant information, which will be subsequently reduced last. In addition, relevance-selection can be done on different modalities to improve the quality of the data reduction process.

Data Enhancement

Data enhancement, driven through data reduction, produces a progressive resolution representation of the initial data. This process enhances the data because no data is lost. Moreover, data are stored in a fashion to allow fast resolution, selection, and retrieval. This process is typically application-driven. For instance, after the creation of consecutive resolutions, the data enhancement process creates a new representation that facilitates constant frame-rate display, or view-dependant display. The fact that the data reduction process can be user-driven allows control of the data enhancement process.

Data Modeling